Research map / Reinforcement learning / Visual and representation RL
Bootstrapped Representations in Reinforcement Learning
Charline Le Lan, Stephen Tu, Mark Rowland, Anna Harutyunyan, Rishabh Agarwal, Marc G Bellemare, Will Dabney
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Pretrain DQN representations with auxiliary cumulant-value predictions | DQN · Four Rooms | online performance | Improved |
| Pretrain DQN representations with auxiliary cumulant-value predictions | DQN · Sparse Mountain Car | online performance | Improved |
| Use ExactSVD or CCR cumulants for offline pretraining | DQN · Four Rooms and Sparse Mountain Car | online performance | Improved |
| Use indicator cumulants instead of isotropic random cumulants (ablation) | supervised auxiliary training · Four Rooms | subspace recovery error | Hurt |
| Use orthogonal cumulant columns (ablation) | supervised auxiliary training · Four Rooms | subspace recovery error | Improved |
| Use standard Gaussian rather than unit-norm Gaussian cumulants (ablation) | TD auxiliary training · Four Rooms | subspace recovery error | Hurt |
| Normalize Gaussian cumulant columns to unit norm (ablation) | TD auxiliary training · Four Rooms | subspace recovery error | Improved |
| Use indicator cumulants rather than Gaussian cumulants (ablation) | TD auxiliary training · Four Rooms | subspace recovery error | Improved |
| Use orthogonal cumulant columns (ablation) | TD auxiliary training · Four Rooms | subspace recovery error | Improved |
| Use isotropic rather than non-isotropic cumulant distributions (ablation) | TD auxiliary training · Four Rooms | subspace recovery error | No clear effect |
Closest papers by meaning
- Bootstrap Latent-Predictive Representations for Multitask Reinforcement Learning (ICML 2020 · similarity 0.89)
- Bridging State and History Representations: Understanding Self-Predictive RL (ICLR 2024 · similarity 0.88)
- A Generalized Bootstrap Target for Value-Learning, Efficiently Combining Value and Feature Predictions (AAAI 2022 · similarity 0.88)
- Behavior Prior Representation learning for Offline Reinforcement Learning (ICLR 2023 · similarity 0.88)
- Reinforcement Learning with Automated Auxiliary Loss Search (NeurIPS 2022 · similarity 0.88)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
